Automatic detection method and system for coronary artery dissections

Abstract

Disclosed in the present invention are an automatic detection method and system for coronary artery dissections. The method comprises: first, collecting clinical coronary artery angiography images, and processing the images to form a data set, so as to train and verify a YOLOv8 model; processing a coronary artery angiography image needing to be tested, and then inputting same into the YOLOv8 model for the identification and localization of an FL in the image; using a constructed SO(2)-UNet model to perform feature extraction and semantic segmentation on coronary arteries in a located region of the image after identification and localization, and accurately segmenting coronary vessels; and further identifying a dissection region in the segmented coronary vessels, and visually presenting same. In the present invention, YOLOv8 is used to perform object detection, a large number of images are quickly and accurately tested in a short time, and coronary artery dissections in the images are automatically identified to realize an automatic detection process, thereby significantly improving the efficiency and accuracy of diagnosis.

Description

Automatic detection method and system for coronary dissection TECHNICAL FIELD

[0001] The present application belongs to the field of image recognition, and particularly relates to an automatic detection method and system for coronary dissection. BACKGROUND

[0002] Coronary dissection is a common cardiovascular disease caused by the intima and media of the coronary artery tearing, forming an intramural hematoma, and the intramural hematoma extending to the outside of the intima, forming a dissection.

[0003] Coronary dissection can occur in different parts such as the aorta, left coronary artery, and right coronary artery. Aortic dissection can be spontaneous or secondary to trauma, aortic disease, etc. If the aortic dissection involves the ascending aorta, it may form aortic valve insufficiency or cardiac tamponade, etc. If it involves the descending aorta, it may form aortic valve regurgitation or cardiac tamponade, etc. Patients usually have symptoms such as chest pain, dyspnea, pallor, sweating, and decreased blood pressure, and in severe cases, shock, etc.

[0004] Coronary dissection can be evaluated by electrocardiogram, echocardiogram, computed tomography angiography, coronary angiography, and radionuclide myocardial perfusion imaging. Although electrocardiogram is a simple examination method, its sensitivity and specificity for coronary dissection are limited, and it may not provide clear diagnostic information in the early or local situation of coronary dissection. Echocardiography can evaluate the structure and function of the heart, but its accuracy is low in detecting post-sternal coronary dissection or rare arterial branch dissection. Although computed tomography angiography can provide high-resolution vascular images, it requires the injection of contrast agents, and there is a risk of radiation exposure. And for some patients, such as those with impaired renal function or those allergic to iodine agents, CTA may not be suitable. Angiography, as a widely used imaging method in vascular surgery, has limitations such as underestimating the true vessel size, plaque morphology, presence of calcium and thrombus, plaque vulnerability, true lesion length, stent expansion and juxtaposition, post-intervention residual stenosis, and the presence of dissection. SUMMARY

[0005] The purpose of the present application is to provide an automatic detection method and system for coronary dissection.

[0006] This system combines the advanced deep learning models of YOLOv8 and SO(2)-UNet to achieve efficient and accurate detection of coronary dissections. YOLOv8, as a target detection model, can quickly locate the dissection regions in images, greatly improving the detection efficiency. Its real-time performance and multi-target detection capability make it particularly suitable for processing large amounts of medical image data. At the same time, the improved backbone network and advanced loss function of YOLOv8 help to improve the accuracy of detection.

[0007] The SO(2)-UNet model plays a key role in blood vessel segmentation. By introducing SO(2) convolution, this model can better capture the anatomical structure of the coronary artery and improve the ability to retain vessel details. This improvement makes the segmentation results more accurate and reduces the situation of boundary blur. SO(2)-UNet combines the multi-scale feature extraction capability of UNet and the rotational invariance of SO(2) convolution, making it very suitable for processing coronary artery images with complex structures. The combination of these two models not only improves the accuracy of detection and segmentation, but also greatly enhances the overall performance and reliability of the system.

[0008] To achieve the above purpose, the present application adopts the following technical scheme: An automatic detection method for coronary dissection, comprising the following steps:

[0009] (1) Collecting clinical coronary angiography images, processing the images to form a data set to train and verify the YOLOv8 model;

[0010] (2) After processing the coronary angiography images that need to be detected, input them into the YOLOv8 model to identify and locate the FL in the images;

[0011] (3) Using the constructed SO(2)-UNet model to extract features and perform semantic segmentation on the coronary arteries in the located area, accurately segmenting the coronary vessels;

[0012] (4) Further identifying the dissection area from the segmented coronary vessels and visualizing it.

[0013] Further, the processing of the coronary angiography images in step (1) specifically adopts the following steps:

[0014] 1.1 Using 3DSlicer or Labelme to label the coronary dissection in the coronary angiography images;

[0015] 1.2 Preprocessing the labeled coronary angiography images by denoising, contrast enhancement, and size standardization.

[0016] Further, the processing of the coronary angiography images in step (2) includes denoising, contrast enhancement, and size standardization.

[0017] Further, the YOLOv8 model in step (2) is fine-tuned with the weight of the dataset before use.

[0018] Further, the SO(2)-UNet model in step (3) uses the isometry module in the E(2)-CNN library.

[0019] Further, in step (4), the further identified interlayer region is specifically subjected to edge detection, contour analysis, spline curve fitting and distance transformation interpolation on the segmented coronary vessels in sequence.

[0020] An automatic detection system for coronary artery dissection, comprising:

[0021] A data acquisition and model training module for acquiring clinical coronary angiography images, processing the images to form a dataset for training and verifying a YOLOv8 model;

[0022] An FL recognition and positioning module for processing the coronary angiography images to be detected and inputting the images into the YOLOv8 model to recognize and position the FL in the images;

[0023] A coronary vessel segmentation module for extracting features and performing semantic segmentation on the coronary arteries in the positioned region using the constructed SO(2)-UNet model to accurately segment the coronary vessels;

[0024] An identification and visualization module for further identifying the interlayer region of the segmented coronary vessels and visualizing the same.

[0025] Further, the data acquisition and model training module includes a data processing submodule for processing the coronary angiography images, which specifically performs:

[0026] 1.1 Labeling the coronary artery dissection in the coronary angiography images using 3DSlicer or Labelme;

[0027] 1.2 Preprocessing the labeled coronary angiography images by denoising, contrast enhancement and size standardization.

[0028] Further, the FL recognition and positioning module further includes a preprocessing submodule for processing the coronary angiography images to be detected, which specifically performs denoising, contrast enhancement and size standardization.

[0029] Further, the identification and visualization module includes an identification submodule and a visualization submodule;

[0030] The identification submodule performs edge detection, contour analysis, spline curve fitting and distance transformation interpolation on the segmented coronary vessels.

[0031] The visualization submodule is used to visually present the identified dissection region.

[0032] The present application has the following beneficial effects:

[0033] 1. Object detection using YOLOv8 can quickly and accurately detect a large number of images in a short time, automatically identify coronary dissection in images, and realize an automated detection process, greatly improving the efficiency and accuracy of diagnosis.

[0034] 2. The system can accurately locate the dissection in the specific anatomical region of the coronary artery, providing more detailed diagnostic information to help doctors develop more precise treatment plans.

[0035] 3. Using a deep learning model, the system has adaptive learning capabilities, constantly optimizing the model with large amounts of data to make it more intelligent and accurate in coronary dissection detection.

[0036] 4. The SO(2)-UNet model better captures the obvious periodic structure of the coronary artery, effectively segmenting the arterial lumen and wall, and improving the accuracy of segmentation.

[0037] 5. Combining UNet structure and SO(2) convolution, it better preserves the details of the blood vessels, reduces the occurrence of artifacts or boundary blurring, and accurately delineates the arterial lumen and wall, making the segmentation results more accurate and reliable.

[0038] 6. Semantic segmentation effectively enhances the image quality of the coronary artery, allowing doctors to more clearly observe the structure and characteristics of the artery, which helps more accurate diagnosis and treatment planning.

[0039] 7. The automatic detection method and system of this system have strong scalability and can be customized and optimized according to actual needs and different medical scenarios, suitable for different types of coronary dissection detection tasks.

[0040] 8. Automatic segmentation of coronary artery lumen and wall can reduce manual operation, improve work efficiency, and reduce the risk of human error, making medical image analysis more reliable and reproducible. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a flowchart of the use of the embodiment coronary dissection automatic detection system.

[0042] Figure 2 is a flowchart of data preparation and preprocessing in the embodiment.

[0043] Figure 3 is a flowchart of FL detection based on YOLOv5s in the embodiment.

[0044] FIG. 4 is an embodiment blood vessel segmentation flowchart.

[0045] FIG. 5 is an embodiment dissection and visualization flowchart.

[0046] FIG. 6 is an embodiment system integration and testing flowchart.

[0047] FIG. 7 is an embodiment physician workflow flowchart.

[0048] FIG. 8 is an embodiment system integration with hospital information system (HIS) flowchart. DETAILED DESCRIPTION

[0049] In order to better understand the technical scheme of the present application, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0050] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0051] Example 1: As shown in FIGS. 1, 2

[0052] Data collection: Collect digital subtraction angiography (DSA) images of coronary artery disease collected in clinical practice. These images are used to train and validate the model to ensure that the model can handle real-world image data.

[0053] Image annotation: Use specialized software tools, such as 3DSlicer or Labelme, to manually annotate the coronary dissection in the DSA images. Annotation includes identifying and marking the boundaries of the false lumen (FL).

[0054] Image preprocessing: Preprocess the images to improve the efficiency and accuracy of subsequent processing steps. Preprocessing steps can include denoising, contrast enhancement, size standardization, etc.

[0055] Example 2: As shown in FIG. 3

[0056] Model selection: Select YOLOv5s as the base model.

[0057] Model Fine-tuning: Fine-tune a pre-trained YOLOv5s model on a manually annotated DSA image dataset. The fine-tuning process involves adjusting the model's weights to improve detection accuracy for FL.

[0058] FL Identification: Apply the fine-tuned YOLOv5s model to new DSA images to identify and locate FLs within the images. The model will output the location and confidence score of the FLs.

[0059] Example 3: As shown in Figure 4

[0060] Model Construction: Construct a SO(2)-Equivariant-UNet model that utilizes equivariant modules from the E(2)-CNN library, ensuring rotational equivariance and invariance.

[0061] Feature Extraction: Extract features from preprocessed DSA images using the SO(2)-Equivariant-UNet model, which utilizes R2Conv layers for group-wise convolution.

[0062] Segmentation Mapping: Combine low-level and high-level features through upsampling and skip connections to generate a rotationally equivariant segmentation map, accurately segmenting the coronary vessels.

[0063] Example 4: As shown in Figure 5

[0064] Edge Detection: Use edge detection algorithms such as the Sobel or Canny operator to identify the edges of blood vessels and dissections in the DSA images.

[0065] Contour Analysis: Perform contour analysis on the detected edges to determine the approximate location and shape of the dissection.

[0066] Spline Curve Fitting: Use spline curve fitting techniques to smooth the contours of the dissection, resulting in more accurate dissection boundaries.

[0067] Distance Transform Interpolation: Apply distance transform interpolation methods to refine the dissection region, improving the accuracy and readability of the segmentation.

[0068] Visualization: Present the segmentation results in a visual manner, such as highlighting the dissection region in different colors on the DSA image, providing intuitive guidance for surgical planning.

[0069] Example 5: As shown in Figure 6

[0070] System Integration: Integrate the above steps into a unified system that can demonstrate the operation process and results on a user interface.

[0071] System Testing: Test the system's performance on an internal validation set, evaluating metrics such as accuracy, recall, F1 score, and others.

[0072] Clinical Evaluation: Collaborate with clinicians to conduct clinical evaluations of the system, collect feedback, and optimize system performance.

[0073] Example 6: As shown in Figure 7

[0074] System Login: The doctor accesses the automatic detection platform through the hospital's network system and logs in using a personal account. The system ensures the security and privacy of all data.

[0075] Image Upload: The doctor uploads the patient's DSA image data. These images may come from a digital subtraction angiography (DSA) machine and are usually stored in DICOM format.

[0076] Image Management: The system provides a user-friendly interface that allows the doctor to view the patient image list, select specific cases, and submit images for analysis.

[0077] Automatic Detection: After submitting the images, the system automatically sends the images to the YOLOv5s-based FL detection module and the SO(2)-Equivariant-UNet blood vessel segmentation module for processing.

[0078] Result Viewing: After processing is complete, the system generates a report including the automatic detection results of FL and the blood vessel segmentation map. The doctor can view these results on the interface.

[0079] Result Evaluation: The doctor evaluates the accuracy of the automatic detection results. The system provides tools to zoom in, zoom out, and pan the image to facilitate the doctor's detailed examination.

[0080] Manual Adjustment: If necessary, the doctor can manually adjust the detection results, such as correcting the boundaries of FL or marking other important blood vessel structures.

[0081] Risk Assessment: Based on the detected FL and other blood vessel structures, the system provides risk assessment tools to help the doctor judge the difficulty and potential risks of the operation.

[0082] Surgery Planning: The doctor uses the automatic detection results and risk assessment to plan the PCI operation, determining the catheter entry point, path, and necessary interventions.

[0083] Surgery Simulation: The system provides surgery simulation functions, allowing the doctor to pre-visualize the operation process in a virtual environment and optimize the operation steps.

[0084] Document Recording: The doctor can generate detailed surgical planning documents, including images, automatic detection results, risk assessments, and operation steps.

[0085] Postoperative Evaluation: After surgery, doctors can re-upload postoperative DSA images, and the system will provide postoperative evaluation tools to help doctors analyze the effectiveness of the surgery.

[0086] Feedback Provision: Doctors can provide feedback on the system's performance, including detection accuracy, user interface experience, and overall satisfaction.

[0087] System Updates: Based on doctors' feedback and new clinical data, the system is continuously updated and optimized to improve detection accuracy and user experience.

[0088] Example 7: As shown in Figure 8

[0089] Integrated Interface: The system integrates with the hospital information system (HIS) through a secure API interface, enabling seamless transmission of patient information and medical image data.

[0090] Patient Information Synchronization: After doctors input patient information in the HIS, the system automatically synchronizes these information to the automatic detection platform, eliminating the need for repeated input.

[0091] Image Automatic Acquisition: The system automatically acquires DSA images related to the patient from the HIS, reducing the need for manual upload.

[0092] Result Sharing: Detection results and surgical planning documents can be automatically sent back to the HIS for access and reference by other medical team members.

[0093] Cross-department Collaboration: The system supports cross-department collaboration, such as information sharing and discussion between cardiology, radiology, and surgery departments.

[0094] Note:

[0095] 1. When training YOLOv8 and SO(2)-UNet models, sufficient sample data and annotations are required to ensure that the models have good generalization ability.

[0096] 2. The model needs to be fine-tuned and optimized to improve the accuracy and efficiency of detection and segmentation.

[0097] 3. In practical applications, factors such as real-time performance, stability, and scalability of the system need to be considered.

[0098] The above detailed the preferred embodiments of the present application. It should be understood that those skilled in the art can make many modifications and changes without creative labor based on the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment based on the existing technology according to the concept of the present application shall be within the protection scope determined by the claims.

Claims

1. An automatic method of detecting a coronary dissection, characterized in that The method comprises the following steps: (1) Collecting clinical coronary angiography images, processing the images to form a data set to train and verify the YOLOv8 model; (2) Processing the coronary angiography images to be detected and inputting the images into the YOLOv8 model to identify and locate the FL in the images; (3) Using the constructed SO(2)-UNet model to perform feature extraction and semantic segmentation on the coronary artery in the located region, and accurately segmenting the coronary blood vessels; (4) Further identifying the FL region from the segmented coronary blood vessels and visualizing the FL region.

2. The method of claim 1, wherein: The processing of the coronary angiography images in step (1) specifically comprises the following steps: 1.1 Labeling the coronary artery dissection in the coronary angiography images using 3DSlicer or Labelme; 1.2 Preprocessing the labeled coronary angiography images by denoising, contrast enhancement, and size standardization.

3. The method of claim 1, wherein: The processing of the coronary angiography images in step (2) comprises denoising, contrast enhancement, and size standardization.

4. The method of claim 1, wherein: The YOLOv8 model in step (2) is fine-tuned with the weight of the model before use.

5. The method of claim 1, wherein: The SO(2)-UNet model in step (3) uses the equivariant module in the E(2)-CNN library.

6. The method of claim 1, wherein: The further identification of the FL region in step (4) specifically comprises edge detection, contour analysis, spline curve fitting, and distance transform interpolation on the segmented coronary blood vessels.

7. An automated detection system for coronary dissection, characterized by It comprises: a data acquisition and model training module for collecting clinical coronary angiography images, processing the images to form a data set to train and verify the YOLOv8 model; an FL identification and positioning module for processing the coronary angiography images to be detected and inputting the images into the YOLOv8 model to identify and locate the FL in the images; a coronary blood vessel segmentation module for using the constructed SO(2)-UNet model to perform feature extraction and semantic segmentation on the coronary artery in the located region, and accurately segmenting the coronary blood vessels; an identification and visualization module for further identifying the FL region from the segmented coronary blood vessels and visualizing the FL region.

8. The system for automatic detection of coronary dissection of claim 7, wherein: The data acquisition and model training module comprises a data processing submodule for processing the coronary angiography images, which specifically performs: 1.1 Labeling the coronary artery dissection in the coronary angiography images using 3DSlicer or Labelme; 1.2 Preprocessing the labeled coronary angiography images by denoising, contrast enhancement, and size standardization.

9. The system for automatic detection of coronary dissection of claim 7, wherein: The FL identification and positioning module further comprises a preprocessing submodule for processing the coronary angiography images to be detected, which specifically performs denoising, contrast enhancement, and size standardization.

10. The system for automatic detection of coronary dissection of claim 7, wherein: The identification and visualization module comprises an identification submodule and a visualization submodule; The identification submodule performs edge detection, contour analysis, spline curve fitting, and distance transform interpolation on the segmented coronary blood vessels; The visualization submodule is used to visualize the identified FL region.

Images